Introducing disambiguating contexts: Pilot 2

Pilot 2 was designed to evaluate the material for the RT study and to test for the accept- ability of Int vs Ext in context. This was done by creating contexts that preceded the target sentence of the type in (1). This context manipulation requires some explanation:

7_{For information on this software see: http://onexp.textstrukturen.uni-goettingen.de/}

8_{The t-test was thus non-standard. It did not test the choice of Int vs. Ext but rather Int or Ext}

The assumption that we make here is that context guides interpretation in that it clearly disambiguates towards one or another reading.

Pilot 2 pursued a twofold goal. (i) It was designed to show that the contexts do their job, i.e. that they disambiguate the interpretation of the comparative as intended. (ii) Reacting to the outcome of Pilot 1, we wanted to test whether the unambiguously internal reading would get higher acceptability ratings than the unambiguously external reading. We elaborate on these two sub-goals briefly: Concerning (ii), we refer to Pilot 1, where we found a preference for the internal compared to the external reading when ambiguous comparatives were presented without context. We interpret this preference for the internal reading as a preference for the less complex interpretation. Adding a disambiguating context should have a strong influence on how the comparative is inter- preted, i.e. it might weaken the effect seen in Pilot 1. However, the context should not influence the difference in complexity between the two readings once disambiguated. We therefore expected that the simpler Int is still easier than the more complex Ext and we thus hypothesized that the simpler internal reading would be judged more acceptable than the external reading (we pick up this hypothesis again in Section 3.3.3 below). To test this hypothesis, we provide the target sentences with contexts that match the targets in the conditions Int/+Match and Ext/+Match.

As to (i), we want to gain evidence that readers actually adopt the internal or the external reading after the respective context. To this end we added conditions with a mismatching context given the induced readings in the conditions Int/–Match and Ext/–Match. The simple idea behind this is that matching contexts will be judged substantially more acceptable than mismatching contexts. Since we conceive of the mis- matching contexts as control conditions, we tested a smaller number of mismatching contexts than matching contexts.

Method

Materials: We used the 24 target sentences from Pilot 1 and constructed four contexts for each of the targets, exemplified in (13). The target sentences from Pilot 1 were supplied with a continuation after the standard of comparison (cf. ‘und darüber freut er sich’ in (13)). These supplements will serve as a spillover region in the RT study reported below in Section 3.3.3.

(13) Peter Peter

traf

met [ein-e[one-f.

stärkere stronger Boxer-in] boxer-f.acc.] als than Julia Julia und and darüber thereabout freut rejoices er he sich. himself

‘Peter met a stronger boxer than Julia and he is happy about it.’

a. Int/+Match: Gestern prahlte Julia vor ihrem Kumpel Peter, sie habe 5 Boxkämpfe mit einem Knockout gewonnen. Peter erinnerte sich, dass eine Boxerin, die er neulich getroffen hat, bis jetzt sogar schon 9 Boxkämpfe durch ein Knockout gewonnen hat.

‘Yesterday, Julia bragged in front of her buddy Peter that she had won 5 boxing matches by knockout. Peter remembered that a female boxer whom he recently met had already won 9 boxing matches by knockout.’

b. Ext/+Match : Gestern erzählte Julia ihrem Kumpel Peter, sie habe sich mit einer Freundin getroffen, die für 5 Knockouts verantwortlich war. Peter erinnerte sich, dass eine Boxerin, die er neulich getroffen hat, sogar schon 9 Knockouts verzeichnen konnte.

‘Yesterday, Julia told her buddy Peter that she met with a friend of hers who was responsible for 5 knockouts. Peter remembered that a female boxer whom he recently met, could even claim 9 knockouts for herself.’

c. Int/–Match: Gestern prahlte Julia vor ihrem Kumpel Peter, sie habe schon 9 Boxkämpfe mit einem Knockout gewonnen. Peter erinnerte sich, dass die stärkste Boxerin, die er je getroffen hat, bis jetzt nur 5 Boxkämpfe gewonnen hat.

‘Yesterday, Julia bragged in front of her buddy Peter that she had won 9 boxing matches by knockout. Peter remembered that the strongest boxer that he had ever met had only won 5 boxing matches so far.’

d. Ext/–Match: Gestern erzählte Julia ihrem Kumpel Peter, sie habe sich mit einer Freundin getroffen, die für ganze 9 Knockouts verantwortlich war. Peter erinnerte sich, dass die beste Boxerin, die er bis jetzt getroffen hat, lediglich 5 Boxkämpfe mit einem Knockout gewonnen hat.

‘Yesterday, Julia told her buddy Peter that she met a friend of hers who was responsible for as many as 9 knockouts. Peter remembered that the best boxer that he had met previously, had only won 5 boxing matches by knockout.’

The context in (13-a), Int/+Match, induces the internal reading of the comparative: Peter met a stronger boxer than Julia is. The context matches the target because the female boxer Peter met won 9 matches whereas Julia won 5 matches. The context in (13-b), Ext/+Match, induces the external reading of the comparative: Peter met a better boxer than Julia met. The context matches the target because the boxer Peter

met won 9 matches whereas the boxer Julia met won 5 matches.

The mismatching context of the control of the internal reading in (13-c), Int/–Match, induces an internal reading and mismatches the target because the boxer Peter met won 5 matches whereas Julia won 9 matches. The mismatching context of the control with the external reading in (13-d), Ext/–Match, induces an external reading and mismatches the target because the boxer Peter met won 5 matches whereas the boxer Julia met won 9 matches.

Since we wanted our subjects to judge no more than 12 experimental items, the total set of 24 experimental items was split into two halves. In order to test twice as many matching than mismatching contexts, we implemented a Latin square design for either half of items, pretending that there were six conditions. The four item variants of the twelve items in a subset were assigned to six lists in a way that eight items on a list instan- tiated the two matching conditions (4 × Int/+Match and 4 × Ext/+Match) and four items instantiated the mismatching controls (2 × Int/–Match and 2 × Ext/–Match). Participants and Procedure: 85 native German speakers took part in the exper- iment, for a payment of ¤5. Six participants were excluded from the analysis because they did not complete the experiment. The remaining participants had a mean age of 27.2 years, ranging from 20 to 63 years; 50 of them were female. The analysis is based on 79 participants and 24 items.

Participants judged the sentences in the given context on a 4-point Likert scale. The experiment was conducted online with OnExp. Figure 3.1 shows how the display of a trial looked like.

The instructions following the context and the target sentence printed in bold read as follows in English:

“Please evaluate the last utterance in bold according to its acceptability in the given context from 1 to 4. 4 means that the utterance is completely acceptable. 1 means that it is not at all acceptable. An utterance is acceptable when it makes sense in the given context and when it is formulated in a way in which a native speaker would use it.”

Results and discussion In order to test whether the contextually induced internal read- ing is judged more acceptable than the contextually induced external reading, we com- pared the mean acceptability ratings for matching contexts: 3.08 for Int/+Match and 3.12 for Ext/+Match. As is already evident from the two means, there is no difference in acceptability between the two readings, as confirmed by the t-test: t1(78) and t2(23)

<1.

I conclude that there is no difference in acceptability of contextually disambiguated comparatives corresponding to the interpretation preference observed in Pilot 1 for am- biguous comparatives presented without context. In particular, the simpler internal reading is not judged more acceptable than the more complex external reading. We will see in the next section whether an on-line measure like reading times reveals effects of the difference in complexity which were not visible in the off-line measure used in Pilot 2, namely acceptability judgments.

There is, however, evidence that the contexts did their job: While items with match- ing contexts were judged acceptable, 3.08 and 3.12 for the internal and the external reading, the controls with mismatching contexts, Int/–Match and Ext/–Match, were considered inacceptable, 1.13 and 1.16.

Referring to our sub-goal (i), the context manipulation, Pilot 2 thus succeeded. Con- cerning sub-goal (ii), the context seems to have a much more important role to play: the structural differences between the two readings seem to be overwritten. I will discuss this point in more detail later.